AI is raising the cost of hospital cyberattacks. Rubrik is pushing a resilience mandate
Cyberattacks aren't just IT incidents anymore. They're care delivery events. As outages stretch from days to months, clinicians stall, patients wait, and staff burn out. Many teams don't come back after living through one.
That's the blunt reality shared by Josh Howell, healthcare CTO at Rubrik, in a recent interview. His message: healthcare needs an enterprise-wide cyber resilience capability, not just more security tools.
The stakes: patient safety and silent failure
AI has amplified both the speed and the scale of risk. Systems don't have to go down for harm to occur. Bad data, misrouted actions, or misaligned agents can quietly create unsafe situations.
Howell pointed to independent safety testing that now lists AI as a top technology threat to patients. There are already hundreds of documented cases where AI contributed to harm, including deaths. That's not abstract. It demands tighter oversight and faster detection.
- ECRI has highlighted AI-related hazards in patient care
- HHS 405(d) HICP outlines industry-validated practices for healthcare cyber risk reduction
Why AI changes your risk model
AI often lands in service lines or research groups before IT sees it. That creates gaps: unknown tools, shadow integrations, and agents with more access than anyone realizes.
Peter Drucker once called healthcare the most complex human organization. Add decentralized AI, and even strong teams struggle to answer basic questions: What data is being touched? Which credentials are in play? What actions can an agent take? How do we validate outputs at scale?
Rubrik's angle: visibility, control, and change discipline
Rubrik is focusing on two fundamentals with its Agent Cloud approach: know what your AI touches, and control how it changes. You can't protect what you can't see, and you can't trust what you don't govern.
A practical example: if your RAG corpus shifts without notice, clinical or operational answers can drift. That shows up as inconsistent decisions, subtle safety issues, and hard-to-trace errors. Tight change control for data, models, and prompts is table stakes now.
Practical playbook for CISOs, CMIOs, and operations leaders
- Inventory AI everywhere: Catalog models, agents, apps, and shadow pilots. Tag owners, use cases, and where each runs.
- Map data exposure: For each AI tool, list data sources, PHI flows, and output destinations. Block unsanctioned connections.
- Tighten credentials: Enforce least privilege, short-lived tokens, and per-agent secrets. Monitor for unusual access patterns.
- Lock down change control: Treat model versions, prompts, RAG corpora, and integrations like code. Require approvals and rollbacks.
- Validate outputs at scale: Define "acceptable" responses. Use sampling, guardrails, and automated checks for clinical and operational tasks.
- Detect silent failure: Build canaries and business-level monitors (orders placed, messages routed, codes called). Don't rely only on uptime.
- Segment and contain: Isolate AI services. Limit blast radius with network and identity boundaries.
- Assume breach: Keep immutable backups of critical data and configurations. Drill recovery by workflow, not just by system.
- Govern vendors: Require transparency on data handling, model updates, and incident response. Bake SLAs into contracts.
- Train the front lines: Give clinical and ops leaders clear playbooks for AI use, escalation, and shutdown triggers.
What this means for your organization
AI can speed care delivery and cut waste, but it also expands your attack surface and error surface. The fix isn't more hype. It's visibility, guardrails, and muscle memory when something goes sideways.
If your teams need structured upskilling on AI risk, governance, and workflows, see curated programs by role at Complete AI Training.
Bottom line
Hospitals can't afford months-long recoveries, staff attrition, or patient harm from silent AI failures. Treat cyber resilience as a system of care. Know what your agents touch, control how they change, and validate what they do-every day, at scale.
Your membership also unlocks:
